Hi Tim,
Our approach is a bit different. We first try to establish whether the
R-free set is biased, by checking whether R-free is surprisingly low
compared to R given the data parameter ratio. If this is the case (or if we
chose a new R-free set for some reason, e.g. because it was too small)
PDB_REDO resets the B-factors to 0.5*(Wilson B-factor) and then does more
cycles of refinement to ensure it converges. This should get rid of the
bias. We compared this to the 'perturb coordinates' approach and in most
cases there wasn't any difference. In some cases the perturbed coordinates
were outside the radius of convergence of Refmac (the version 5.2 or perhaps
5.4) particularly in cases with NCS. So coordinate perturbation was just not
worth it. This was before NCS restraints were properly implemented in
PDB_REDO (hurray for local NCS/LSSR!), so this issue must be much smaller
now.
Of course the rebuilding round of PDB_REDO followed by more refinemnt, will
cause enough model perturbation if the above results are not convincing
enough ;)
We also do full (well, k-fold) cross-validation for small data sets in which
the different test sets are all-but-one completely biased. Here, too, the
B-factor resetting works well enough. That said we might add a few extra
cycles of refinement here to be on the safe side.
Cheers,
Robbie
> -----Original Message-----
> From: CCP4 bulletin board [mailto:[log in to unmask]] On Behalf Of
> Tim Gruene
> Sent: Thursday, February 28, 2013 10:33
> To: [log in to unmask]
> Subject: Re: [ccp4bb] Rfree flag
>
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Dear Kavya,
>
> As far as I understand the PDBRedo project attempts to make the
reflections
> unbiased from the structure by a random shift of coordinates (e.g. 'NOISE'
> keyword in pdbset, although I am not aware of an investigation about
> whether this actually does make remove bias.
>
> It is safest to keep the same Rfree-set.
>
> Regards,
> Tim
>
> On 02/28/2013 06:54 AM, Kavyashree Manjunath wrote:
> > Dear users,y
> >
> > Is it mandatory to use the same reflections for Rfree calculations of
> > a ligand bound data as that of its native?
> >
> > Thank you With Regards Kavya
> >
> >
>
> - --
> - --
> Dr Tim Gruene
> Institut fuer anorganische Chemie
> Tammannstr. 4
> D-37077 Goettingen
>
> GPG Key ID = A46BEE1A
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.12 (GNU/Linux)
> Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
>
> iD8DBQFRLyQ2UxlJ7aRr7hoRAsHBAJ43K7f2lcSZwm6fD1pH8+grvrOqRACg1bdi
> PUUKJaUv8C4JgmcPNM6H9+U=
> =J47i
> -----END PGP SIGNATURE-----
|